Will the winner get everything in the world of robots?

Today, several dozens of companies are trying to create robots technology - these are the OEMs, their traditional suppliers, the existing large technology companies and start-ups. Obviously, not everyone is destined to succeed, but a sufficiently large number of them have every chance - so it will be curious to think about what the consequences of the “winner gets everything” effect will be and what leverage can be in this area. Will there be a network effect , due to which one or two of the largest companies will squeeze out all the others, as happened in the world of smartphones or PC operating systems? Or is there a place on the market for five to ten companies that will compete for a very long time? And what kind of floors in this pyramid victory will bring power over other layers?

Such questions are quite important, because they point to the balance of power in the auto industry of the future. The world in which automakers can buy a product like turnkey autonomy from any of the five to six companies (or make it yourself), just like they buy ABS technology today, is very different from the world in which Waymo and Uber was probably the only real candidate capable of building a business model on its own, as Google did with Android. Microsoft and Intel found pain points in the PC world, and Google found it in smartphones; What can be such points at robomobil?

It is immediately clear that the goods will be equipment and sensors for autonomy. There are quite a lot of engineering efforts and scientific research in them, just like, say, in LCD screens, but there is no reason to use one instead of the other just because everyone else does. The effects of economies of scale are quite strong, but there is no network effect. So, let's say, LIDAR will go from a “rotating bucket from KFC” worth $ 50,000, to a small widget without moving parts worth several hundred dollars, and there will be winners in this segment - but there’s no network effect, because the LIDAR winner will give you no more leverage on other floors of the pyramid (unless you can capture a monopoly) than Sony's best photo matrices (which she sells to Apple) give her in the world of smartphones. In the same way, batteries (and motors and battery / motor control) will be the same commodities that RAM serves today - economies of scale, scientific research, and probably several winners in each category, but without much leverage.
')
On the other hand, it is unlikely that there will be direct parallels with the ecosystems of third-party software developers, such as they currently exist for PCs or smartphones. Windows squeezed out a Mac, and then iOS, and Android squeezed out Windows Phone thanks to a vicious circle tied to developers, but you won’t buy a car based on how many applications you can run on it. Uber, Lyft and Didi will work for them all, and all of them will have Netflix built into the screens, but other applications will work on your phone (or clock, or glasses).

You need to search not in the cars themselves, but above the pyramid - in standalone software, which allows the car to move along the road, without crashing into anything, in optimization in the city scale and in building routes, during which we can automate all the machines in the system, but not only each separately - and on the wave of all this there will be parks from robotaxi. Network effects in the field of on-demand cars are obvious, but with the advent of autonomy, they will become very complicated (autonomy will cut the cost of any on-demand trip by three quarters, or even more). Robotaxi parks will dynamically redistribute their cars, as well as coordinate among themselves, and probably together with all the machines at once, their routes in real time to achieve the greatest efficiency, in order to avoid situations in which all cars simultaneously choose one route. This, in turn, can be combined not only with fluctuating pricing, but also with path-based pricing — you may need to pay more to get to a place at peak times faster, or you can choose your arrival time depending on cost.

From a technical point of view, these three floors (movement, paving and optimization, exit on request) are quite independent of each other - presumably, it will be possible to install Lyft application in GM GM and allow the Waymo module installed in the system to drive the car. Obviously, someone is hoping for the appearance of levers of pressure on different levels, or perhaps combining them into one package - Tesla plan to forbid people to use their robobili with other travel services on request, except for their own. In the other direction it does not work –Uber will not insist on using only its autonomous system. But although Microsoft used joint leverage in promoting Office and Windows, both of these positions won first place in their own markets as part of their own network effects: if a small OEM insists on using its small robotaxi service, it will look as if In 1995, Apple insisted on buying AppleWorks instead of Microsoft Office. I suspect that the result will be a more neutral approach. Especially if we have long-distance car traffic coordination, or even direct communication between cars at intersections — some kind of common pyramid floor will be needed here (although I feel susceptible to decentralized systems).

All this, of course, is a complete theorizing, and looks like an attempt to predict today's traffic jams while living in 1900. The only area in which we can talk about key network effects is autonomy itself. It all depends on the equipment, sensors, software, but primarily on the data. And in the case of autonomy, two kinds of data are important - maps and driving data. Let's start with the cards.

Our brain constantly processes data from the senses and builds a three-dimensional model of the world around us, in real time and at the level of the subconscious - when we run through the forest, we usually do not stumble on the roots or bang our head on the branch. In the world of autonomy, this is called SLAM - Simultaneous Localization and Mapping (Method of Simultaneous Localization and Mapping). We are engaged in marking our environment and localizing ourselves within it. This is obviously a basic requirement for autonomy - the ro-mobile must understand where it is on the road, what may surround it (rows, turns, sidewalks, traffic lights, etc.), and it must understand where other vehicles are and how fast they move.

So far, the implementation of this technology on a real road in real time is a rather difficult task. People drive with the help of sight (and sound), but building a fairly accurate three-dimensional model of your environment based on only one image (especially two-dimensional) remains an unsolved problem: machine learning makes this possible, but nobody has yet achieved the accuracy required for driving. Therefore, we use workarounds. Therefore, almost all stand-alone projects combine image processing with LIDAR at 360 degrees: each sensor has its limitations, but their combination (“sensor fusion”) helps you to build a complete picture. Building the world around you using images alone will probably be possible at some point in the future, but using a large number of sensors allows you to speed up this process, even if you have to wait until the cost and form factor of these sensors are reduced to practical ones. LIDAR is a workaround for building the world around you. Once this is obtained, then machine learning is often used to understand what this world is like - whether this form is a car, a cyclist. But in this case, the network effect is not observed - you can gather up quite a lot of images of cyclists, even without having a fleet of cars.

If LIDAR is one of the SLAM workarounds, the other, more interesting, is pre-built maps, that is, high-resolution 3D models. You drive around the road in advance, calmly process all the data, build models of the streets, and then place them in any car that drives them. RoboMob no longer needs to process all this data, and turning out turns or lights from the rest of the noise in the real world at a speed of 65 miles per hour — instead, he knows where the traffic light is located and can locate himself on key milestones roads at any time. Therefore, your car uses cameras and LIDAR to determine its position on the road, search for traffic lights, etc., comparing what it can see with a previously created map, and it does not need to do this on its own and from scratch. He also uses this data to recognize other vehicles in real time.

Cards have a network effect. When any robot moves along a previously marked road, it both compares the road with the map and updates the map. Each robot can also be a sightseeing car at the same time. If you have sold 500,000 robots, and someone else has 10,000, your cards will be updated more often and more accurately, so the chance to meet something completely new and unexpected and get confused by your cars will be less. The more cars you sell, the better they behave - the network effect, by definition.

The risk of this is that in the long run, all the machines will be able to implement SLAM without LIDAR, and drive without pre-installed cards - after all, people cope with this. Whether it will happen at all, and when it is, is not yet clear, but now it seems that it will happen not immediately after the first ro-mobiles are on sale, and by that time everything will change.

If the cards are the first network effect in the data, then the second is derived from what the machine does after it understands its environment. Driving on an empty road or on a road filled with other mobile vehicles is one task, but determining what other people are going to do on the road and how to handle it is a completely different task.

One of the breakthroughs that support autonomy is that machine learning should be well suited for this: instead of trying to write down complex rules explaining what you think about how people can behave, MO will use data, and more is better. The more data about the behavior and reactions of real drivers in the real world you can collect (both the behavior of other drivers and the behavior of the drivers of your own survey cars), the better your software will understand what is happening around and the better it will plan next steps. As in the case of maps, prior to launch, your test cars collect this data, but after starting, each car you sell also collects this data and sends it home. So just like in the case of cards, the more cars you sell, the better you will have them - again the network effect is by definition.

Driving data can be used in another place - in the simulation. This should answer the questions “if X happens, how will our standalone software behave?” One of the ways to answer it is to make a mobile and send it on a trip around the city to see how it reacts to the random actions of other drivers. The problem is that such an experiment is not controlled - it is impossible to return to the same situation with the new software and see what is changing and whether the problems have been fixed. Accordingly, a lot of efforts are now being made to create simulations - put your robot software in Grand Theft Auto (almost literally) and check how you like. This does not necessarily help you catch all the options for events (“will LIDAR determine the presence of this truck on the road?”), And some simulation scenarios will be looped, but it will tell you how your system will behave in certain situations, and you can collect these situations from travel data in the real world. So the indirect network effect works here: the more data you have about traveling in the real world, the more accurate your simulations will be, and, consequently, the better your software will be. The simulations also show obvious advantages depending on the scale - how many people will work on it, how many computer resources you can devote to it, and what theoretical experience do you have in working on large computing projects. The fact that Waymo is part of Google clearly gives it an advantage: its robobomies reel 25,000 real miles every week, and in 2016, they drove an average of 19 million miles per week in simulations.

We can say that Tesla has an advantage both in terms of maps and driving data: from the end of 2016, those cars whose owners bought the autopilot supplement have eight cameras each that give a 360-degree view supported by a radar directed forward (and there is also a set of ultrasonic sensors that work at a fairly small distance and are mainly used when parking). All this can collect both data on cartography and driver behavior, and send them to Tesla - and, apparently, recently the company really began to collect such data. The catch is that, since the radar is only forward, Tesla will have to use only image processing to build models of most of the surrounding world, but, as I noted, we still do not know how to do this with sufficient accuracy. This means that Tesla, in fact, collects data that no one today is able to read (or, at least, read well enough to come up with a complete solution). Of course, this problem will need to be addressed both for data collection and for the car itself, so Tesla makes an uncharacteristic bet on the rapid development of computer vision. Tesla saves time without waiting for the appearance of cheap / practical LIDAR (today Tesla would not be able to place them on all its cars), but without LIDAR software for computer vision will have to solve more complex tasks, so this can also take a long time. And if all the other parts of the software necessary for autonomy - parts that decide what the machine should do in a certain situation - will not appear soon, during this time LIDAR can become cheaper and more practical, and the Tesla workaround will lose its meaning. We'll see.

So, network effects — the winner gets everything — are in the data: in the driving data and in the maps. This raises two questions: who will get this data, and how much do we need it?

Possession of data is an interesting question of power and value. Obviously, Tesla plans to independently create all the important parts of the technology and implement it in its own cars, so that it owns the data. But some OEMs claimed that if the machine belongs to them, and the customer is theirs, then the data also belongs to them, and not to any other technological partners. This is a fairly reasonable position, which should be considered in connection with the manufacturers of sensors: I am not sure that it will be possible to sell GPUs, cameras or lidars by themselves, and not share the available data with anyone. But a company that produces robomobils needs to have data — nothing will work out without them. If changes to the data do not affect the technology within the context of a neat connection, the technology will not improve. This means that OEMs increase the network value in favor of the supplier, and they themselves have nothing from this value, except for some improvement in autonomy - but this improved autonomy itself becomes a commodity among all products of any OEM that uses it. This is similar to the position of PC or Android: they create a network effect, agreeing to use software in their products, thanks to which they sell their products, but their product has become a commodity, and the network value goes to the tech company. This is a vicious circle in which the value for the most part goes to the seller, not to the OEM. Therefore, most OEMs now want to make robobom on their own - they don’t want to finish just like Compaq .

And this leads me to the final question: how much data you may need? Can the system infinitely improve with the constant addition of data, or will we observe an S-shaped curve - will there be some point after which the addition of new data will have very little effect on the improvements?

That is, how strong is the network effect?

Pretty obvious question for maps. For the maps to become good enough, what density of cars and what frequency of trips do you need, and what is the smallest market share that results? How many participants can accommodate the market? Can there be a dozen companies, or just two? Can a bunch of second-rate OEMs get together and merge all their data into one database? Will trucks be able to sell their data in the way they sell various cartographic information today? The situation is different from the consumer software ecosystems - RIM and Nokia could not merge the Blackberry and S60 user bases together, but the cards can be merged.Is this a barrier to entry or condition for entry to the market?

This question applies to the data needed for travel, and in general to all projects using MO: at which point the improvements become insignificant when adding new data, at what point does this curve begin to straighten, and how many people are needed to get this amount of data? For example, for general-purpose search engines, the effect of improvement seems endless - the answer can almost always be more relevant. But for autonomy, it seems necessary to have a ceiling - if the car can drive around Naples for a year and not get confused, what else can be improved there? At some point, you essentially finish the improvements. The network effect means that the product improves with an increase in the number of users - but how many users will be required for the product to stop improving significantly? How many cars need to sell,to get your autonomy closer to the best on the market? How many companies can achieve this? And, in the meantime, the MO itself is changing rapidly - one cannot exclude the possibility that the amount of data required to achieve autonomy will decrease dramatically.

In all this lies the assumption of the very existence of such concepts as better or worse autonomy. What does “worst” autonomy mean? Slightly more likely to die in an accident, or a little more likely that the car will get tangled, slow down at the curb and connect with the support center so that the operator takes control? Will manual controls, packed in polyethylene, jump out of the console, and will the machine give encouraging comments?

– , ?
– , , , , – . – . . , .
.
– ! – . – -… .
– ? – .
- Not. ?
– .
– ?
– .
– , – . – .
– , – .
– , – . – ! !
– , – .
, , , , , . .
.
– , , – , – ! -…
– , ! – . – .
. , . : .

The answer, apparently, is that the fifth level of autonomy will appear in the form of an evolution of the fourth level - each machine will have manual controls, but they will be used less and less, and the fifth level will appear step by step, and the controls will decrease, then hide, and then disappear altogether - atrophy. Perhaps, first the 5th level will appear in Germany, then in Naples, then in Moscow [ dirty insinuations - approx. trans. ]. This will mean that data is collected on a network scale and used long before the appearance of complete autonomy.


[ “Rebuilding in Moscow,” the tweet author comments. It is unlikely that this glaring case can be called the norm. For comparison, it’s enough to google “usa road wars” - approx. trans. ]

The answers to these questions are unknown to us. Few experts in this field expect a level of autonomy of the fifth level for five years, most of them lean towards ten years. However, they point to a wide range of results that can lead to extremely different options for influencing the automotive industry.

One of the extremes - the network effect will be weak, with the result that 5-10 companies with a more or less autonomous platform will appear on the market. In this case, the automaker will buy autonomy as a component, at a price similar to ABS, airbags or satellite navigation. The industry will still change - autonomy will lead to a drop in the cost of travel on request by at least three-quarters of the current price, with the result that many people will think about the need to own their own car. At the same time, the transition to electric cars will reduce the number of moving parts in a car by 5-10 times, which will drastically change the engineering dynamics, the supplier base and barriers to entry to the market. But the situation does not reach the level of Android.

At the other extreme, only Waymo will succeed in creating a robotic car, in which case the industry will look completely different.